.webp){kind=icon}
In recent years, artificial intelligence (AI) has experienced phenomenal growth and is disrupting the global economy. Large American companies such as OpenAI, Microsoft or Nvidia have invested billions of dollars in it. However, a Chinese startup named DeepSeek comes to upset this landscape by offering open source language models (LLM) high-performance, developed at a lower cost.
In addition, the recent announcement of the merging the digital activities of TechTarget and Informa Tech reinforces the debate around the creation of global technological ecosystems. In this article, we will explore in detail the rise of DeepSeek, the reasons for its success, its differences with OpenAI, as well as the importance of TechTarget/Informa Tech consolidation in the AI race.
What is DeepSeek?
DeepSeek is an AI research laboratory based in Hangzhou (China), founded in May 2023 by Liang Wenfeng, a graduate of Zhejiang University. Supported by the investment fund High-Flyer, DeepSeek's mission is to democratize AI by developing open source models at low production costs.
- Founding date : May 2023
- Head office : Hangzhou, China
- Vision : Open and simplify access to major language models (LLM)
- Financing : Of private origin, not publicly disclosed
With an approach resolutely focused on Open source And the collaboration, DeepSeek aims to compete with American tech giants, including OpenAI, by proving that excellence in AI does not necessarily require colossal financial resources or the most advanced chips.
The meteoric success of DeepSeek
On January 20, 2025, DeepSeek launched DeepSeek R1, a language model that focuses on advanced reasoning. Its impact was immediate:
- Mobile app : The DeepSeek AI assistant quickly became the most downloaded app on the Apple App Store, surpassing ChatGPT.
- Announced development costs : Less than 6 million dollars, while competing models can cost several hundred million.
- Open source : DeepSeek R1 is offered under an open license, making it easy for developers around the world to adopt and integrate.
This resounding success caused panic in financial markets, especially in the United States, where investors worried about the profitability in the long term, AI giants that have invested heavily in proprietary solutions.
DeepSeek vs. OpenAI: Key differences
Despite common points (global ambition, use of cutting-edge language models), DeepSeek and OpenAI diverge strongly on several aspects.
This Cost difference And the openness policy have contributed to the explosion of the popularity of DeepSeek, which attracts many freelance developers, startups and companies looking for a cheaper alternative to American solutions.
The key innovations of DeepSeek R1
The DeepSeek R1 stands out for its ability to treat complex reasoning tasks. Several technological points explain this advance:
- Reinforcement Learning (RL) : Instead of relying primarily on supervised learning, DeepSeek favors a large-scale approach through reinforcement.
- Reward Engineering : The reward system is codefined by rules (rule-based) rather than by a classical neural model, for a best performance in reasoning.
- Distillation : DeepSeek researchers were able to compress knowledge into models as small as 1.5 billion parameters while maintaining most of the functionalities.
- Emergent Behavior Network : Discovery of a emerging behavior derived from reinforcement learning, propelling creativity and the ability to solve complex problems without explicit programming.
The different DeepSeek language models
Since its creation in 2023, DeepSeek has already released several models:
- DeepSeek Coder (Nov. 2023)
- First open source model specially dedicated to programming tasks.
- DeepSeek LLM (Dec. 2023)
- General-purpose model for conversations and text generation.
- DeepSeek-v2 (May 2024)
- Increased performance and reduced training costs.
- DeepSeek-Coder-V2 (Jul. 2024)
- 236 billion parameters, 128,000 token context window, tailored for advanced coding challenges.
- DeepSeek-v3 (Dec. 2024)
- Mixture-of-experts architecture, 671 billion parameters, 128,000 token context window, covering a wide range of tasks.
- DeepSeek-R1 (Jan. 2025)
- Based on V3, focused on the advanced reasoning. Direct competition from OpenAI gpt-O1, but at much lower cost.
- Janus-Pro-7B (Jan. 2025)
- Vision model capable of understand and generate images, paving the way for new creative perspectives.
Why is DeepSeek worried about the United States
The success of DeepSeek created a shock wave, especially the January 27, 2025, when NASDAQ fell 3.4% at the open. Some companies like Nvidia have lost up to 17%, or around $600 billion in market valuation. Several factors explain these concerns:
- Cost disruption : DeepSeek is developing its models for only a few million dollars, threatening the profitability of historical players.
- Limiting chip exports : Although China is experiencing restrictions on American GPUs and AI accelerators, DeepSeek proves that it is possible to reach a very high technological level without the latest chips.
- Open source model : The free and open nature of DeepSeek democratize access to AI, in direct contradiction to the commercial strategies of many American firms.
- Geopolitics : Seen as a new “Sputnik Moment” by some investors, the success of DeepSeek highlights the rise of China in the AI race.
DeepSeek in the face of cyberattacks
The January 27, 2025, at the height of its popularity, DeepSeek suffered a major malicious attack. The company had to temporarily limit the registration of new users, but the service for existing accounts remained operational. Speculations point to a DDoS attack targeting its APIs and its web platform.
In addition, the January 29, 2025, Wiz Research has unveiled a security flaw in a DeepSeek database displaying sensitive information (API keys, chat histories, logs). DeepSeek reacted by deactivating the database as soon as the alert was received, but the duration of exposure is not clearly known.
TechTarget/Informa Tech fusion: a new reference center
Along with the rise of DeepSeek, the B2B information and services industry is evolving. TechTarget, a leader in technological media, and Informa Tech, a major player in IT events and training, have announced the merger of their digital activities.
The objectives of this merger
- Consolidation of the offer : Bringing together TechTarget's media expertise and Informa Tech's event and data capabilities.
- Accelerating digital transformation : Offer businesses more comprehensive services to help them navigate the AI era.
- International reach : Create a unique ecosystem at the service of IT decision makers around the world.
The impact on the industry
With this merger, we are witnessing the emergence of a new center of reference in the world of digital services for AI and big data. This contributes to:
- Reinforcing competition against American and Chinese platforms.
- Encourage the dissemination of good practices in terms of AI and security.
- Supporting innovations in content editing, data analysis, and IT talent training.
Conclusion: The future of AI and tech ecosystems
The DeepSeek case perfectly illustrates the volatility And the speed to which artificial intelligence is progressing. In just two years, this Chinese startup has developed high-performance models, Open and above all much less expensive than those of its competitors. The American giants must now adapt to this new situation.
At the same time, the TechTarget/Informa Tech merger demonstrates the desire to create integrated ecosystems around Technology watch, of media And happenings. The objective? Helping businesses better understand and adopt these AI innovations.
As the AI race intensifies, the international collaboration, the cybersecurity And thequick adaptation to new technologies will be decisive. DeepSeek, with its open source positioning, could well play the role of catalyst for a next wave of AI solutions that are more accessible and less expensive.
